OPTIMAL PORTFOLIOS IN DEFINED CONTRIBUTION PENSION SYSTEMS by Eduardo Walker Professor School of Business Pontificia Universidad Católica de Chile September 2005 I appreciate the comments to previous versions of this work by the participants to the following seminars: “Investment Round Table by Fidelity” (Villarrica, Chile, October 2003); research seminar of Facultad de Ciencias Económicas y Administrativas (Reñaca, Chile, January 2004), particularly Felipe Zurita; Latin America Econometric Society (July 2004). The opinions herein only represent the author’s. Email: [email protected] ; Address: Vicuña Mackenna 4860, Santiago, Chile. Phone: (56 2) 354 4002; Fax: (56 2) 553 1672 OPTIMAL PORTFOLIOS IN DEFINED CONTRIBUTION PENSION SYSTEMS ABSTRACT We study optimal portfolios for defined contribution (possibly mandatory) pension systems, which maximize expected pensions subject to a risk level. By explicitly considering the present value of future individual contributions and changing the risk-return numeraire to future pension units we obtain interesting insights, consistent with the literature, in a simpler context. Results naturally imply that the local indexed (inflation-adjusted) currency is the benchmark and that the investment horizon is long. Optimal portfolios have a hedging component with an even longer duration than a deferred (real) pension, which begins its lifetime payments upon retirement. Results are illustrated with the parameters obtained for the United States by Campbell and Viceira (2001). Keywords: optimal portfolio, defined contribution, pension reform, deferred pension, emerging markets JEL: G11, G18, G23 2 1. INTRODUCTION We study optimal portfolios for defined contribution (possibly mandatory) pension systems. Specifically, we consider the perspective of an individual pension account holder who wishes to maximize her expected pension level subject to a certain risk level. We do not consider whether market incentives eventually lead a pension system to provide some kind of optimal portfolio. We neither explicitly consider the interrelationship that may exist between current consumption- savings-portfolio decisions and pension fund decisions, which affect consumption upon retirement, eventually assuming that these decisions are considered independently. This could be justified in the context of “mental accounting” (Thaler (1985)), but more simply, by considering that in some pension systems people are institutionally forced to save for retirement in a single-purpose individual pension fund account. Naturally, in a Modigliani Miller spirit, fully rational individuals could partially or completely undo in their own portfolios whatever decision is made in the delegated pension fund portfolio. Therefore, results presented here may be relevant for those who cannot undo these decisions. The question thus is what constitutes an optimal asset allocation from a long-term perspective, considering the available funds in the individual account and a sequence of future contributions to it. As such, this problem has not been solved, at least stated in this way, but closely related problems have been solved before and some of our results turn out to be applications or particular cases. However, from the analysis presented here some interesting and new results arise. A (presumably novel) way of summarizing the problem is to restate it by changing the numeraire: both expected returns and risk have to be measured in real future pension units. This guides our analysis. Thus, the benchmark obviously becomes the local inflation-adjusted currency and the horizon is long-term. From the perspective of a future pensioner, short-term instruments are not risk free since they have low correlation with the future cost of a pension. For hedging purposes, we are interested in assets whose returns are highly correlated with variations in future pension costs, and these naturally are long-term local indexed bonds, although certain kinds of equity possibly have good hedging properties too. In any case, taking as reference a deferred (real) pension, which begins lifetime payments upon retirement, an important result is that for moderate levels of risk aversion, the optimal portfolio has an even longer duration. From a very different perspective, it is important to consider that in some cases pension funds manage mandatory savings with minimum pension government guarantees. The cost of such a guarantee is directly related with the risk of obtaining low pensions. This reinforces the idea of a conservative asset allocation bias, as the one suggested here. 3 2. LITERATURE REVIEW Many of the models developed in the literature assume a constant short term real interest rate, eliminating by assumption reinvestment risk. Among these we find Samuelson and Merton (1969). Assuming complete markets and constant relative risk aversion they show that the fraction invested in the riskless asset is constant and independent of age and wealth, depending only on expected returns, variances and risk aversion. Merton (1971) also considers nonrandom labor income and finds that the optimal fraction of wealth invested in the risky asset is a function of its risk premium and the relative importance of the present value of labor income. A larger fraction is invested in the risky assets the larger the relative importance of the present value of labor income. Bodie, Merton and Samuelson (1992) and Jagannatan and Wang (1996) ratify that human capital is an important determinant of the portfolio decision. Bodie et al show that considering future labor income has two effects: a wealth effect (as in Merton (1971)) and a substitution effect, through which riskier human capital implies less invested in risky assets. They also find that investors whose labor supply is more elastic are willing to take more risk. Campbell, Cocco, Gomes y Maenhout (1999) show that higher labor income volatility and higher correlation between shocks to labor income and asset returns imply less investment in risky assets. Viceira (2001) adds that the wealth effect can also be extended to the fraction of labor income uncorrelated with the risky asset returns, and Wang (2002), that what really matters are the volatility and correlation of the unpredictable parts of labor income. Cocco, Gomes y Maenhout (2001) find that an important determinant of the optimal portfolio is the ratio of cumulative wealth to expected labor income, which is non-stationary during the lifecycle. Empirical studies by Faig and Shum (2000) and Cocco (2000) tend to confirm these predictions, and the behavior of other nonfinancial assets affects investment decisions in a way that is similar to human capital. Good part of the previous studies, by assuming a fixed short-term real rate, does not consider an important element of defined contribution pension (mandatory) systems: the volatility in the cost of the final pension. With random interest rates both the investment horizon and reinvestment risk become important. If no inflation-indexed bonds exist, we might get Siegel’s (1998) conclusion, that stocks dominate bonds in the long run. However, this is not a general conclusion. Final pension volatility depends on accumulated wealth but also on the cost of buying an annuity upon retirement. This means that we have to add long-term real interest rate risk to that of the history of pension fund investments. This leads Campbell and Viceira (2002, chapter 3) to consider particularly appropriate for a long term investor to invest in long-term indexed bonds.i Bajeux-Besnainou, Jordan and Portait (2003) show that the relevant risk free portfolio is a 4 combination of long and short-term bonds, which depends on the investment horizon. For example, if a constant duration long-term bond fund exists, the investor will optimally combine this fund first with short sales of the short-term asset and eventually, with long positions in it, depending on the remaining investment horizon. They also find that an individual with constant relative risk aversion will keep a fixed fraction of her wealth invested in her own riskless portfolio. Finally, Bodie (2002) quite reasonably argues that a person’s welfare must be judged from the perspective of lifetime consumption and leisure and not just final wealth. This is studied formally in Bodie at al (2004), who allow for habit formation, flexible labor supply, stochastic wages and opportunity set, and liquidity constraints. The bottom line of these studies is that a person should consider all of her wealth (financial and nonfinancial) during her entire lifetime to determine the optimal portfolio decision. The question is what lessons can be learned from this literature review for the portfolio problem of a defined contribution (possibly mandatory) pension system. First, given that the mandate seems clear (to provide the best possible pensions with the minimum possible risk, given a sequence of contributions to the individual account) it is evident that we cannot consider only final wealth as the argument in a welfare function. Second, we must consider somehow the fact that there will be future contributions to the account, which will have to be invested. Third, the argument for recommending more equity investment for younger people does not assume (wrongly) that equity is safer in the long run. Younger people can compensate any loss in the individual pension account by slightly increasing savings, although this may be a somewhat questionable assumption in the context of a mandatory pension system.